Dark confinement and chiral phase transitions: gravitational waves vs matter representations

Page created by Harvey Rhodes
 
CONTINUE READING
Published for SISSA by        Springer
                                                                         Received: October 10, 2021
                                                                        Accepted: December 17, 2021
                                                                         Published: January 3, 2022

Dark confinement and chiral phase transitions:
gravitational waves vs matter representations

                                                                                                      JHEP01(2022)003
Manuel Reichert,a,1 Francesco Sannino,b,c,d,2 Zhi-Wei Wangc,e,3 and Chen Zhangf,4
a
  Department of Physics and Astronomy, University of Sussex,
  Brighton, BN1 9QH, U.K.
b
  Scuola Superiore Meridionale,
  Largo S. Marcellino, 10, 80138 Napoli NA, Italy
c
  CP3 -Origins, University of Southern Denmark,
  Campusvej 55, 5230 Odense M, Denmark
d
  Dipartimento di Fisica “E. Pancini”, Università di Napoli Federico II | INFN sezione di Napoli,
  Complesso Universitario di Monte S. Angelo Edificio 6, via Cintia, 80126 Napoli, Italy
e
  Department of Astronomy and Theoretical Physics, Lund University,
  22100 Lund, Sweden
f
  INFN Sezione di Firenze,
  Via G. Sansone 1, I-50019 Sesto Fiorentino, Italy
    E-mail: m.reichert@sussex.ac.uk, sannino@cp3.sdu.dk, wang@cp3.sdu.dk,
    chen.zhang@fi.infn.it
Abstract: We study the gravitational-wave signal stemming from strongly coupled mod-
els featuring both, dark chiral and confinement phase transitions. We therefore identify
strongly coupled theories that can feature a first-order phase transition. Employing the
Polyakov-Nambu-Jona-Lasinio model, we focus our attention on SU(3) Yang-Mills theories
featuring fermions in fundamental, adjoint, and two-index symmetric representations. We
discover that for the gravitational-wave signals analysis, there are significant differences
between the various representations. Interestingly we also observe that the two-index sym-
metric representation leads to the strongest first-order phase transition and therefore to a
higher chance of being detected by the Big Bang Observer experiment. Our study of the
confinement and chiral phase transitions is further applicable to extensions of the Standard
Model featuring composite dynamics.

Keywords: Cosmology of Theories beyond the SM, Thermal Field Theory, Confinement,
Spontaneous Symmetry Breaking

ArXiv ePrint: 2109.11552
    1
      ORCID: https://orcid.org/0000-0003-0736-5726.
    2
      ORCID: https://orcid.org/0000-0003-2361-5326.
    3
      Corresponding author. ORCID: https://orcid.org/0000-0002-5602-6897.
    4
      Corresponding author. ORCID: https://orcid.org/0000-0003-2649-8508.

Open Access, c The Authors.
                                                          https://doi.org/10.1007/JHEP01(2022)003
Article funded by SCOAP3 .
Contents

1 Introduction                                                     2

2 Effective theories for dark gauge-fermion sectors                3
  2.1 Basic considerations                                         3
  2.2 The PNJL model                                               5
  2.3 Polyakov-loop potential                                      7

                                                                       JHEP01(2022)003
  2.4 Condensate energy                                            9
  2.5 Zero-point energy and medium potential                      10
  2.6 Model parameters and observables                            14
  2.7 Comments on universality                                    15

3 Confinement and chiral phase transitions                        16
  3.1 Cosmological considerations                                 16
  3.2 Order parameters                                            17
  3.3 Bubble nucleation                                           19
  3.4 Gravitational-wave parameters                               21
      3.4.1 Inverse duration time                                 21
      3.4.2 Energy budget                                         22
      3.4.3 Bubble-wall velocity                                  23
      3.4.4 Efficiency factors                                    23
  3.5 Gravitational-wave spectrum                                 24

4 Results                                                         24
  4.1 Choice of PNJL parameters                                   25
  4.2 Parameter scan                                              26
  4.3 Gravitational-wave spectrum                                 28
  4.4 Signal-to-noise ratio                                       30

5 Discussions                                                     32
  5.1 Thin-wall approximation                                     32
  5.2 Models with cubic terms in the condensate energy            33

6 Conclusions                                                     34

A Wave-function renormalization                                   35

B Cosmological and laboratory constraints on massive dark pions   42

                                      –1–
1    Introduction

Future gravitational-wave (GW) observatories provide new opportunities to investigate the
existence of dark sectors that are currently inaccessible. Because the experiments will be
sensitive to strong first-order phase transitions in the early universe, it is paramount to
understand the landscape of theories that can lead to GW signals. Within the landscape of
theories, asymptotically free gauge-fermion systems are privileged, being already chosen by
nature to constitute the backbone of the Standard Model (SM). It is therefore reasonable
to expect them also to appear in the dark sector [1–14].
     In [15], we embarked on a systematic investigation of the composite landscape by

                                                                                                  JHEP01(2022)003
providing the first comprehensive study of the dark confinement phase transition stemming
from pure gluonic theories. There, we first extended the state-of-the-art knowledge of
the confinement phase transition to arbitrary number of colours by combining effective
approaches with lattice results and then determined their GW imprints.
     Using our work [15] on pure gauge dynamics as a stepping stone, we now investigate the
dark dynamics stemming from the confinement and chiral phase transitions arising when
adding fermions in different matter representations to gauge theories. For previous works
and complementary approaches on chiral and confinement phase transitions see e.g. [16–28].
We highlight below the main findings of our work:

    1. We have systematically investigated the chiral and confinement phase transitions in-
       cluding their interplay for SU(3) Yang-Mills theory with fermions in fundamental,
       adjoint, and two-index symmetric representations via the Polyakov-Nambu-Jona-
       Lasinio (PNJL) model.

    2. Representations matter: fermions in the two-index symmetric representation increase
       the strength of the first-order confinement phase transition while the fermions in the
       adjoint representation decrease it. Notably, for the adjoint and two-index symmetric
       representation cases with one Dirac flavour, the chiral phase transition is a second-
       order phase transition.

    3. For all representations (chiral and confinement phase transitions), the inverse dura-
       tion of the phase transition is large, β̃ ∼ O(104 ). We discuss the large value of β̃ in
       the thin-wall approximation where it is given by a competition of the surface tension
       with the latent heat. This sheds light on generic features of non-abelian gauge-fermion
       systems and helps finding models with stronger gravitational-wave signals.

    4. The confinement phase transition with fermions in the two-index symmetric repre-
       sentation has the best chance of being detected by the Big Bang Observer (BBO)
       with a potential signal-to-noise ratio SNR ∼ O(10).

    5. We further provide an outlook of which strongly coupled models can potentially lead
       to a first-order chiral phase transition. The methods used in this work can be readily
       applied to some of these models as well as other dark and bright extensions of the
       Standard Model featuring new composite dynamics.

                                             –2–
This work is structured as follows. In section 2, we introduce the PNJL model as
an effective low energy model which we use here to describe the dynamics in the dark
gauge-fermion sector. In section 3, we discuss the details and interplay of the chiral and
confinement phase transitions, including the bubble nucleation and the GW parameters
and spectrum. In section 4, we present our results which includes a detailed analysis of
the GW spectrum for the various models and the corresponding signal-to-noise ratios. In
section 5, we discuss our results in the light of the thin-wall approximation and analyse
which further models are likely to have a first-order phase transition. In section 6, we offer
our conclusions.

                                                                                                 JHEP01(2022)003
2     Effective theories for dark gauge-fermion sectors

2.1    Basic considerations
A first-order phase transition in the early universe generates a GW signal that might be
detectable by future observations [29–40], for recent reviews see, e.g., [41, 42]. This offers
unprecedented detection possibilities for dark sectors that are otherwise (almost) decoupled
from the visible SM sector. In this section, we concentrate on the description of such dark
phase transitions, which fall into two categories:

    1. Confinement phase transition characterized by the restoration of the centre sym-
       metry at low temperatures [43]. The order parameter is the traced Polyakov loop [44].

    2. Chiral phase transition characterized by the spontaneous breaking of chiral sym-
       metry at low temperatures. The order parameter is the chiral condensate.

These phenomena occur in the strongly coupled regime of the gauge dynamics and cannot
be described by perturbation theory. To make further progress, three approaches can be
envisioned:

    1. Universality analysis [45–47]. This is useful for investigating the order of the phase
       transition but does not provide a quantitative way to compute thermodynamic ob-
       servables near a first-order phase transition.

    2. First-principle non-perturbative approaches, e.g. lattice gauge theory [48] and func-
       tional renormalization group [49]. Unlike pure-gauge theories, first-principle results
       for thermodynamic observables are very limited for gauge-fermion theories in the
       chiral limit (i.e. zero fermion mass limit).

    3. Effective theories [50, 51]. These are constructed by including the relevant degrees of
       freedom and enforcing symmetry principles. They allow for a quantitative framework
       to compute thermodynamic observables near a first-order phase transition. However,
       there is the possibility that they do not provide a faithful modelling of the phase
       transition dynamics. The results obtained from effective theories provide valuable
       hints about possible dynamics of the underlying gauge-fermion theory, but should
       always be interpreted with care.

                                             –3–
Fermion                            Centre
 Model name Gauge group             Reality Nf KMT term             Chiral symmetry breaking pattern
                           irrep                           symmetry
    3F3        SU(3)        F         C    3       6F         ∅        SU(3)L × SU(3)R → SU(3)V
                                                  12F
    3G1        SU(3)       Adj        R    1                  Z3         SU(2) × U(1)A → SO(2)∗
                                               (Ignored)
                                                  10F
     3S1       SU(3)        S2        C    1                  ∅         U(1)L × U(1)R → U(1)V ∗
                                               (Ignored)

Table 1.     The three dark gauge-fermion models studied here, see section 2.1 for a detailed
explanation.

                                                                                                       JHEP01(2022)003
In this work, we employ effective theories to obtain a quantitative description of the phase-
transition dynamics in dark gauge-fermion sectors. For the description of the chiral phase
transition at finite temperature, chiral effective theories such as the Nambu-Jona-Lasinio
(NJL) model [52, 53], see [54–57] for reviews, and the Quark-Meson (QM) model [58, 59]
are frequently adopted in the literature. In order to account also for the confinement phase
transition, these models have been generalized to include the Polyakov-loop dynamics,
with quarks propagating in a constant temporal background gauge field. The resulting
models are called Polyakov-Nambu-Jona-Lasinio (PNJL) model [60, 61] and Polyakov-
Quark-Meson (PQM) model [62, 63], respectively. See also [64] for a study of the Polyakov-
extended linear sigma model. Here, we focus on the PNJL approach and leave the PQM
approach for future work.
     We study three dark gauge-fermion models as shown in table 1. In the “Fermion irrep”
column, F, Adj, and S2 denotes the fundamental, adjoint, and two-index symmetric repre-
sentation, respectively. The “Reality” column refers to the reality property of the fermion
representation, which is complex (C) for F and S2 , and real (R) for Adj. The “KMT
term” column indicates the number of fermions needed to form a Kobayashi-Maskawa-’t
Hooft term [65, 66] (i.e. the ’t Hooft determinantal term), e.g. 6F means the KMT term is
a six-fermion interaction. For the 3G1 and 3S1 models the KMT terms are 12-fermion and
10-fermion terms respectively, and in our PNJL treatment, their effects are ignored since
they correspond to operators of very high dimensions. The chiral symmetry breaking pat-
terns for 3G1 and 3S1 models are also marked with an asterisk to indicate that the effects
of KMT term are not considered (so the U(1)A part is included in the chiral symmetry).
     We restrict our attention to the SU(3) gauge group for simplicity. The treatment of
the PNJL grand potential in gauge groups of larger rank requires introducing two or more
independent Polyakov-loop variables and is left for future work. Three smallest fermion
representations F, Adj, and S2 are then the natural targets for investigation. For the
fundamental representation case, we consider the three Dirac flavour case, which is most
likely to exhibit a first-order chiral phase transition (see discussion on universality below).
The phase transition and GW signature of the 3F3 model have been also studied in [19],
using several effective theories including the PNJL model. Compared to [19], we employ
a different regularisation as discussed below. For the adjoint and two-index symmetric
representations, we consider one Dirac flavour since the case of two Dirac flavours is believed

                                               –4–
to be close to the lower boundary of the conformal window [67–69].1 We work in the chiral
limit, i.e. setting all current quark masses to zero for simplicity. In a more generic setup,
nonzero current quark masses could be introduced in such cases, leading to pseudo-Nambu-
Goldstone bosons (pNGB). We leave this more generic case for future study.

2.2     The PNJL model
The PNJL Lagrangian can be generically written as [60, 61]

                                LPNJL = Lpure-gauge + L4F + L6F + Lk ,                               (2.1)

                                                                                                              JHEP01(2022)003
where Lpure-gauge is the pure-gauge part of the Lagrangian whose effect is to contribute as
the Polyakov-loop potential in the full grand potential, to be described below. L4F and
L6F are the multi-fermion interaction terms which exist in the NJL model. They read
                         8
                         X
       3F3 : L4F = GS         [(ψ̄λa ψ)2 + (ψ̄iγ 5 λa ψ)2 ] ,
                        a=0
             L6F = GD [det(ψ̄Li ψRj ) + det(ψ̄Ri ψLj )] ,                                            (2.2)
                        h                                                     i
       3G1 : L4F = GS (ψ̄ψ)2 + (ψ̄iγ 5 ψ)2 + |ψ C P ∗ ψ|2 + |ψ C P ∗ γ 5 ψ|2 ,       L6F = 0 ,       (2.3)
       3S1 : L4F = GS (ψ̄L ψR )(ψ̄R ψL ) ,                                           L6F = 0 ,       (2.4)

Here the same notation ψ for the dark quark fields, and GS for the four-fermion coupling are
adopted in all three models to avoid proliferation of new symbols. Note that ψ is a colour
triplet in 3F3, a colour octet in 3G1, and a colour sextet in 3S1. It is understood that in the
above equations, the colour indices of ψ are contracted to form singlets inside the fermion
bilinear it resides in. For example, ψ̄ and ψ are contracted with a Kronecker delta in colour
space, while ψ C and ψ are contracted with a unitary symmetric matrix P ∗ in the colour
space in the 3G1 case.2 In the 3F3 model case, ψ is also a three-component vector in the
flavour space and we write ψ = (u, d, s)T when we want to make its individual components
explicit. Note that these are dark quarks, not toqbe confused with the SM quarks. The
λa are 3 × 3 matrices in flavour space with λ0 ≡ 23 1 and λa , with a = 1, . . . , 8, are the
usual Gell-Mann matrices written in the flavour space, normalized as Tr F (λa λb ) = 2δ ab
(TrF denotes trace in the flavour space only). Finally, Lk is the covariant kinetic term for
the quark field [60, 61]

                       Lk = ψ̄iγu Dµ ψ ,                        Dµ = ∂ µ − iAµ ,                     (2.5)

with Aµ = δ0µ A0 being the temporal background gauge field living in the corresponding
fermion representation. In the Polyakov gauge [60, 72], A0 can be taken to be diagonal and
   1
    If the two Dirac flavour case is inside the conformal window, then there is no confinement and chiral
symmetry breaking at low temperature. It is also possible that the two Dirac flavour case is just below the
lower boundary of the conformal window, associated with a weakly first-order phase transition in the sense
discussed in [70]. Some nice discussions also appear in [71].
   2 C
    ψ is defined as ψ C ≡ C ψ̄ T as usual, with C being the charge conjugation matrix. The existence of
the unitary symmetric matrix P is guaranteed in the 3G1 case because the adjoint representation is real.

                                                     –5–
static. In the mean-field approximation to be introduced later, A0 is also taken to be spa-
tially homogeneous so it acts as a constant imaginary chemical potential [60] (A0 = −iA4
when continued to Euclidean spacetime). This way of coupling the gauge field to the quark
field allows us to investigate the interplay between confinement and chiral dynamics in a
convenient manner. However, only temporal gauge fields play a role in the modelling here.
It is expected that for high temperatures (a few times the confinement phase-transition
temperature), the transverse gluons are also important and the PNJL modelling should be
revised accordingly [61, 73].
     A few remarks are in order regarding the construction of L4F and L6F in the NJL
model. In principle one should write down to a given order (e.g. four-fermion level) all

                                                                                                   JHEP01(2022)003
operators that are compatible with the full symmetry (spacetime, colour, and flavour) of
the theory, with each independent operator carrying an independent coefficient [74, 75]. A
further complication arises due to the fact that in computing fermion loops, both the direct
term and the exchange term may contribute. One can achieve simplifications by considering
a Fierz-invariant Lagrangian from the beginning, which can be obtained by adding to the
original Lagrangian its Fierz transformation [55, 75]. With the Fierz-invariant Lagrangian,
one only needs to consider the direct terms in the computation [55]. Fortunately, in many
cases (including the present work) we do not need to carry out the exercise of writing down
the full Fierz-invariant Lagrangian compatible with the symmetry of the theory. This is
because one may work in a mean-field approximation and only care about the condensate in
some but not all channels. For example in the mean-field approximation here, we are only
concerned with the condensate hψ̄ψi since we work at finite temperature but zero chemical
potential. We therefore only need to retain four-fermion terms related to this particular
channel and express them in a form that preserves the full symmetry of the theory. Adding
the Fierz transformation amounts to a redefinition of the couplings in the terms relevant
for calculation. Since the coupling GS and GD are left arbitrary at this stage, we may also
stick to the Fierz-non-invariant Lagrangian and take into account only the direct terms in
the computation, without loss of generality [57].
     The chiral symmetry of the L4F term in the 3F3 case can be made manifest by in-
troducing composite fields Φij ≡ ψ̄jR ψiL , (Φ† )ij ≡ ψ̄jL ψiR with i, j = 1, 2, 3 being flavour
indices and making use of the following identity for the four-fermion term [56, 76]
                                          8 h
                                       1X                               i
                        TrF (Φ† Φ) =          (ψ̄λa ψ)2 + (ψ̄iγ 5 λa ψ)2 ,                (2.6)
                                       8 a=0
which can be proven by brute force using the explicit form of the λa matrices. Φ transforms
as (3, 3̄) under the chiral symmetry group SU(3)L × SU(3)R , and thus TrF (Φ† Φ) is invari-
ant under SU(3)L × SU(3)R . The six-fermion term L6F in the 3F3 case is a parametrization
of the U(1)A -breaking instanton effect. It is also chirally-invariant because it can be writ-
ten as GD (detΦ + h.c.) and the SU(3)L and SU(3)R transformation matrices have unit
determinants. ! In the 3G1 model, the L4F term can be rewritten using the Nambu spinor
         ψL
Ψ≡          C  as [77]
        P ψR
                 h                                                 i
      L4F = GS (ψ̄ψ)2 + (ψ̄iγ 5 ψ)2 + |ψ C P ∗ ψ|2 + |ψ C P ∗ γ 5 ψ|2 = GS |ΨC ΣΨ|
                                                                               ~ 2,       (2.7)

                                             –6–
!
                         ψLC                    ~ matrices are defined by Σ1 = 1, Σ2 =
in which   ΨC   ≡ P        C )C and the 2 × 2 Σ
                      (P ψR
iσ1 , Σ3 = iσ3 . The unitary matrix P is a symmetric matrix in colour space such that
P ψRC has the same gauge transformation properties under the gauge group as ψ . This
                                                                                 L
rewriting of L4F using Nambu spinors makes the SU(2) × U(1)A symmetry manifest in the
                                                                        ~ can be shown
3G1 case, since Ψ transforms as a doublet in the Nambu space, while ΨC ΣΨ
                                                              3
to transform as a complex three-vector in the Nambu space. Finally, in the 3S1 case,
the L4F term in eq. (2.4) is already manifestly invariant under the chiral U(1)L × U(1)R
symmetry.
      The finite-temperature grand potential of the PNJL models can be generically writ-

                                                                                                              JHEP01(2022)003
ten as
                                        h       i           h         i           h        i
        VPNJL = VPLM [`, `∗ ] + Vcond hψ̄ψi + Vzero hψ̄ψi + Vmedium hψ̄ψi, `, `∗ .                   (2.8)

Here ` and `∗ denote the traced Polyakov loop and its conjugate (to be defined more
precisely below). VPLM [`, `∗ ] is the Polyakov-loop potential describing the glue sector where
all the
      h potential
           i      can be fully determined by the existing    h lattice
                                                                   i    results (see e.g. [15]).
Vcond hψ̄ψi represents the condensate energy while Vzero hψ̄ψi denotes the fermion zero-
                                                       h                  i
point energy. The medium potential Vmedium hψ̄ψi, `, `∗ encodes the interactions between
the chiral and gauge sector which arises from an integration over the quark fields coupled
to a background gauge field. Each part is described in detail in the following sections.
    Before moving on it is important to note that PNJL models are non-renormalizable
due to the multi-fermion interactions. We truncate the models to the six-fermion operators
at most and use a 3D cutoff to obtain meaningful predictions from divergent momentum
integrals. The cutoff Λ should be understood as a model parameter [55]. This regularization
scheme is most convenient for finite-temperature computation and has been widely used in
the NJL literature.

2.3     Polyakov-loop potential
Pisarski proposed the Polyakov-Loop Model (PLM) in [78, 79] as an effective field theory
to describe the confinement-deconfinement phase transition of SU(N ) gauge theories. The
fundamental (traced) Polyakov loop ` plays the role of the order parameter (Trc denotes
the trace in colour space)
                                                       1
                                            ` (~x) =      Trc [L] ,                                  (2.9)
                                                       Nc
where
                                                    " Z                       #
                                                        1/T
                                 L(~x) = P exp i            A4 (~x, τ ) dτ ,                        (2.10)
                                                        0

   3
     In [77] there is a second term written at the four-fermion level for the NJL Lagrangian of one Dirac
flavour of quark in a real representation of the gauge group. That term explicitly breaks U(1)A (but not to
a correct discrete subgroup) and is mistaken for the ’t Hooft determinental interaction by the authors. We
therefore do not include it.

                                                     –7–
is the thermal Wilson line at temperature T , P denotes the path ordering along the time
direction, and A4 is the Euclidean temporal component of the gauge field in the fundamental
representation (with the gauge coupling absorbed).4 The symbols ~x and τ denote a spatial
point and the Euclidean time, respectively. Note that an ordinary gauge transformation of
the gauge field reads

                          Aµ (x) → A0µ (x) = V (x) [Aµ (x) + i∂µ ] V † (x) ,                           (2.11)

with the transformation matrix V (x) ∈ SU(Nc ). A centre symmetry transformation is
defined to be a transformation in the form of eq. (2.11) with V (x) satisfying a twisted
boundary condition (β ≡ 1/T ) [51, 80]

                                                                                                                JHEP01(2022)003
                  V (x4 = β) = zk · V (x4 = 0) ,                                zk = ei2πk/Nc ,        (2.12)

for k = 0, 1, . . . , Nc − 1. Such a centre symmetry transformation preserves the periodic
boundary condition for Aµ along the time direction. The fundamental Polyakov loop
transforms non-trivially under the ZNc centre symmetry

                       ` → zk ` ,                          k = 0, 1, . . . , Nc − 1 .                  (2.13)

The connection between the centre symmetry and confinement is due to the fact that the
free energy of a single static quark in the fundamental representation Fq is related to the
thermal average of the traced fundamental Polyakov loop [51, 80]

                                             exp(−βFq ) ∝ h`i .                                        (2.14)

So an unbroken centre symmetry implies the vanishing of h`i, which in turn implies Fq
is infinite, and vice versa. Note the above discussion of centre symmetry applies to the
pure-glue sector. The coupling to dynamical fermions may or may not explicitly break
the centre symmetry, depending on the quark representation [80]. For the three models in
table 1, fermions in the fundamental and two-index symmetric representation of the SU(3)
gauge group leads to explicit breaking of the centre symmetry, while fermions in the adjoint
representation preserve the centre symmetry. This is indicated in the “centre symmetry”
column of table 1.
     We adopt the Polyakov gauge [60, 72] in which A4 is diagonal and static. Also, in
the spirit of mean-field approximation, we consider A4 to be spatially homogeneous [81].
Then ` is independent of ~x. Introducing also the conjugate traced Polyakov loop `∗ and
the notation |`|2 ≡ ``∗ , the simplest effective potential preserving the ZNc symmetry in the
polynomial form is given by
                                            b2 (T ) 2
                                                                                             
                       (poly)
                     VPLM = T 4 −                  |`| + b4 |`|4 − b3 `Nc + `∗Nc ,                     (2.15)
                                               2
where
                                                          2                   3
                                            T0      T0                    T0                   T0 4
                                                                                            
                   b2 (T ) = a0 + a1           + a2            + a3                 + a4           .   (2.16)
                                            T       T                     T                    T
   4
    Note that for all three models, 3F3, 3G1 and 3S1, the Polyakov-loop potential is a potential for the
Polyakov loop in the fundamental representation.

                                                    –8–
Parametrization                 a0         a1         a2         a3                a4     b3         b4
                polynomial                      3.72       -5.73      8.49       -9.29             0.27   2.40       4.53
                logarithmic                     4.26       -6.53      22.8       -4.10                    -1.77

       Table 2. Parameters in the effective Polyakov-loop potentials, see eqs. (2.17) and (2.15).

We have chosen the coefficients b3 and b4 to be temperature independent following the
treatment in [15, 51, 61], and also neglected higher orders in |`|.
     For the SU(3) case, there is also an alternative logarithmic parameterization which
includes the information on the Haar measure,5 see e.g. [51, 84], given by

                                                                                                                                           JHEP01(2022)003
                                     a(T ) 2
                                                       h                                i                           
             (3log)
            VPLM      =T   4
                                   −      |`| + b(T ) ln 1 − 6|`|2 + 4(`∗3 + `3 ) − 3|`|4 ,                                       (2.17)
                                       2

with
                                                                     2                   3                             3
                                            T0                  T0                   T0                             T0
                                                                                                            
                 a(T ) = a0 + a1                     + a2                 + a3                 ,   b(T ) = b3                 .   (2.18)
                                            T                   T                    T                              T

                                        (poly)               (3log)
The ai , bi coefficients in VPLM and VPLM have been determined with a dedicated fit
in [15] to available pure-glue lattice data [85]. The results are shown in table 2.
     The temperature T0 in eqs. (2.16), (2.18) is identical to the critical temperature of the
confinement phase transition Tc , in the pure-glue case. In the presence of fermions, the
relation of T0 to Tc is slightly modified, which we will discuss later.

2.4      Condensate energy
                                            h          i
The condensate energy Vcond hψ̄ψi can be viewed as a tree-level contribution from the
chiral condensate hψ̄ψi to the grand potential VPNJL . It can be derived in a self-consistent
mean-field approximation [19, 56, 86] in which one introduces auxiliary fields for the con-
densate and splits the original Lagrangian into a mean-field part LMFA and a residual
interacting part Lres so that in the Bogoliubov-Valatin (BV) vacuum defined by a set of
self-consistent conditions (SCC), Lres has vanishing expectation value, and the SCC coin-
cide with the equations of motion for the auxiliary fields derived from LMFA . For computing
the condensate energy, the procedure can be simplified to a linearization of the fermion
bilinears around the condensate [57, 87]. Moreover, as explained in section 2.2, we only
need to consider direct terms.
     Here we outline the derivation of the condensate energy for the 3F3 case. The only
                                 ¯ and hs̄si. In L4F , the condensate energy then comes
relevant condensates are hūui, hddi,
from the combination

                                                                      ¯ 2 + 2(s̄s)2 .
                      (ψ̄λ0 ψ)2 + (ψ̄λ3 ψ)2 + (ψ̄λ8 ψ)2 = 2(ūu)2 + 2(dd)                                                         (2.19)
   5
    The Haar measure in the context of confinement physics is the Jacobian of the variable transformation
from the gauge potential to the Polyakov loop [82, 83].

                                                                 –9–
Then the trick is to rewrite (ūu)2 as

        (ūu)2 = [(ūu − hūui) + hūui]2 = (ūu − hūui)2 + 2hūui (ūu − hūui) + hūui2
              ' −hūui2 + 2hūuiūu ,                                                        (2.20)

where in the last step the (ūu − hūui)2 term is dropped in the spirit of the mean-field
approximation. In the remaining terms, the 2hūuiūu term contributes to the constituent
quark mass of u which plays an important role in determining the zero-point energy and
medium part of the potential. The −hūui2 term leads to a contribution to the condensate
                                                ¯ 2 and (s̄s)2 , and to L6F . In the chiral
energy. Similar procedures can be applied to (dd)

                                                                                                      JHEP01(2022)003
limit, the condensates should exhibit flavour universality, therefore we introduce, in the
3F3 case

                                              ¯ = hs̄si = 1 hψ̄ψi .
                           3F3 : σ ≡ hūui = hddi                                            (2.21)
                                                          3

The condensate energy is then

                                                      1
                               3F3 : Vcond = 6GS σ 2 + GD σ 3 ,                              (2.22)
                                                      2

which is consistent with [19, 88] after conversion of notations and conventions. The σ 3 term
that originates from the ’t Hooft determinental interaction turns out to be an important
driving force for a first-order chiral phase transition. In the 3G1 and 3S1 cases, we define

                                  3G1 and 3S1 : σ ≡ hψ̄ψi ,                                  (2.23)

and their condensate energies are found to be

                                    3G1 : Vcond = GS σ 2 ,
                                                  1
                                    3S1 : Vcond = GS σ 2 .                                   (2.24)
                                                  4

In these cases, the ’t Hooft determinantal interaction is associated with some very high-
dimensional operator which we neglect in the current approximation, and thus there is no
σ 3 term.

2.5   Zero-point energy and medium potential

In the mean-field approximation outlined in section 2.4, the effects of the multi-fermion
interaction terms L4F and L6F boil down to a contribution to the condensate energy shown
in eqs. (2.22) and (2.24), and a contribution to the constituent quark mass as discussed
below eq. (2.20). With the inclusion of the constituent quark mass, the covariant kinetic
term for the quark field shown in eq. (2.5) becomes

                 L0k = ψ̄(iγu Dµ − M )ψ ,                     D µ = ∂ µ − iAµ ,              (2.25)

                                             – 10 –
with Aµ = δ0µ A0 and the constituent quark mass M is given by
                                                      1
                                3F3 : M = −4GS σ − GD σ 2 ,                             (2.26)
                                                      4
                               3G1 : M = −2GS σ ,                                       (2.27)
                                             1
                                3S1 : M = − GS σ ,                                      (2.28)
                                             2
and the result for the 3F3 case is again found to be consistent with [19, 88] after conversion
of notations and conventions. In the Polyakov gauge and the mean-field approximation,
A0 = −iA4 is diagonal and constant, acting as an imaginary chemical potential. The
contribution of L0k to the grand potential can be readily evaluated by a functional inte-

                                                                                                             JHEP01(2022)003
gration over the fermion at finite temperature and imaginary chemical potential as L0k is
quadratic in the fermion field. The calculation can be found in standard textbooks in
thermal field theory [89]. The resulting contribution to the grand potential hcan be     i de-
composed into a temperature-independent zero-point energy contribution Vzero hψ̄ψi and
a temperature-dependent
        h            i      thermal quark energy (called medium potential) contribution
                   ∗
Vmedium hψ̄ψi, `, ` . The expression for the zero-point energy is given by [51, 57]

                                                                                d3 p
                                          h   i                            Z
                                   Vzero hψ̄ψi = −dim(R) 2Nf                         Ep ,           (2.29)
                                                                               (2π)3
where
                                                             q
                                               Ep =            p~ 2 + M 2 ,                         (2.30)
is the energy of a free quark with constituent mass M and three-momentum p~, dim(R) is
the dimension of the quark representation R, and Nf is the number of Dirac quark flavours.
The momentum integral is understood to be regularized by a sharp three-momentum cutoff
Λ, which enters the expression for observables and is thus also a parameter of the theory.
The integration can be carried out analytically and the result is [50]
                              dim(R)Nf Λ4
                                                                                      p
                                                                ξ4      1 + ξ2 − 1
                h         i                         q                                          
                                                 2
        Vzero       hψ̄ψi = −        2
                                          (2 + ξ   )   1 + ξ2 +    ln p            ,                (2.31)
                                  8π                            2       1 + ξ2 + 1
                     M
in which ξ ≡         Λ.   The spontaneous chiral symmetry breaking at zero
                                                                        h
                                                                           temperature
                                                                              i
                                                                                       is the
result of the interplay between the negative contribution from Vzero hψ̄ψi which favours
                                                                                     h      i
large values of M , and the positive contribution from Vcond hψ̄ψi which favours small
values of M [57].                  h         i
    The medium potential Vmedium hψ̄ψi, `, `∗ is evaluated to be [51]

                                                                            d3 p
                                      h                 i              Z
                                                    ∗                                   †
                              Vmedium hψ̄ψi, `, `           = −2Nf T             (SR + SR ),        (2.32)
                                                                           (2π)3
                 †
in which SR and SR are defined as
                                                            Ep − µ
                                                                                   
                                  SR ≡ TrC ln 1 + LR exp −            ,
                                                              T
                                                            Ep + µ
                                                                 
                                   †
                                  SR ≡ TrC ln 1 + L†R exp −           .                             (2.33)
                                                              T

                                                             – 11 –
Here µ is the chemical potential that we take to be zero. LR is the Polyakov loop matrix
in the quark representation R, defined in a similar manner to eq. (2.10)
                                                  " Z                        #
                                                      1/T
                                                          (R)
                                LR (~x) = P exp i           A4 (~x, τ ) dτ ,                        (2.34)
                                                        0

with the Euclidean temporal gauge field A4 now taken to be in the representation R.
Accordingly, we also define the normalized traced Polyakov loop in the representation R

                                TrC [LR ]                                        TrC [L†R ]
                    `R (~x) =             ,                         `∗R (~x) =              .       (2.35)
                                dim(R)                                           dim(R)

                                                                                                              JHEP01(2022)003
                                      †
We now proceed to evaluate SR and SR    at zero chemical potential. In the spirit of the mean-
field approximation, we utilize the properties of the traced Polyakov loops that are satisfied
at the saddle point of the grand potential. Especially, we note that the traced Polyakov
loop `R at the saddle point is always real,6 and becomes equal to its conjugate `∗R at zero
chemical potential [51]. With this in mind, in the grand potential we set `R = `∗R [19, 90],
and the LR in the fundamental representation of SU(3) can be parameterized as
                                                    n                o
                                       LF = diag eiθ , e−iθ , 1 ,                                   (2.36)

assuming Polyakov gauge and spatial homogeneity. The phase θ is then related to `R in
the fundamental representation as
                                                        3`F − 1
                                              cos θ =           ,                                   (2.37)
                                                           2
          †
SR and SR   in the fundamental representation can now be easily evaluated using the
parametrization in eq. (2.36), and the result is
                           h                                         i                         
                  SF = ln 1 + (3`F − 1)e−Ep /T + e−2Ep /T + ln 1 + e−Ep /T ,                        (2.38)

with SF† = SF . To evaluate SR and SR               †
                                                      in higher representations, we note that SR is
invariant under a similarity transformation in colour R-representation space, thus only
the eigenvalues of LR matter for the calculation. LR can always be brought into a di-
agonal form via a similarity transformation, and its diagonal entries become pure phase
factors exp(iλR                                                                              R
                   j ), j = 1, 2, . . . , dim(R) since LR is unitary. The eigenvalue phases λj , with
j = 1, 2, . . . , dim(R), are weights of the irreducible representation R, and they can be ob-
tained from the weights (i.e. eigenvalue phases) of the fundamental representation which
we parameterize. For example, in the adjoint representation case, LAdj can be parameter-
ized as
                                       n                                            o
                         LAdj = diag e2iθ , e−2iθ , eiθ , e−iθ , eiθ , e−iθ , 1, 1 ,                (2.39)
   6
    This is because the saddle-point value of `R is just the thermal average h`R i in the presence of zero
external source, and thus has the interpretation of the free energy of a static quark in the representation
R. Away from the saddle point, `R does not need to be real.

                                                   – 12 –
with θ defined in the parameterization of LF in eq. (2.36). SR in the adjoint representation
is then computed to be
                                                         
                    SAdj = 2 ln 1 + e−Ep /T
                                        h                                            i
                                + ln 1 + (9`2F − 6`F − 1)e−Ep /T + e−2Ep /T
                                            h                                    i
                                + 2 ln 1 + (3`F − 1)e−Ep /T + e−2Ep /T ,                     (2.40)

      †
with SAdj = SAdj . For the two-index symmetric representation, LS2 can be parameterized as

                                                                                                      JHEP01(2022)003
                                                  n                      o
                                LS2 = diag e2iθ , e−2iθ , 1, 1, eiθ , e−iθ ,                 (2.41)

with the same θ introduced in eq. (2.36). SR in the two-index symmetric representation is
then computed to be
                                                         
                    SS2 = 2 ln 1 + e−Ep /T
                                        h                                            i
                                + ln 1 + (9`2F − 6`F − 1)e−Ep /T + e−2Ep /T
                                        h                                    i
                                + ln 1 + (3`F − 1)e−Ep /T + e−2Ep /T ,                       (2.42)

with SS† 2 = SS2 . The expressions we obtained for SR in eqs. (2.38), (2.40), and (2.42) agree
with [90].
    If we consider a gauge group of larger rank, then the fundamental traced Polyakov
loop `F alone is not sufficient to describe the medium potential. For example, in the SU(4)
case, we need two angles to parameterize LF
                                                      n                  o
                                   (4)
                                 LF = diag eiθ1 , e−iθ1 , eiθ2 , e−iθ2 ,                     (2.43)

where the superscript “(4)” is used to indicate the quantities associated with the SU(4)
gauge group. The θ1 , θ2 angles can be traded for two independent traced Polyakov loops,
in the fundamental and two-index antisymmetric representations, respectively

                                             (4) 1
                                            `F = (cos θ1 + cos θ2 ) ,
                                                 2
                                             (4) 2
                                            `A2 = cos θ1 cos θ2 .                            (2.44)
                                                 3
                          (4)               (4)
Then SR depends on `F and `A2 simultaneously. This is true even for fermions in the
fundamental representation
                      h                                                                  i
              (4)                 (4)                              (4)
             SF = ln 1 + 4`F (e−Ep /T + e−3Ep /T ) + 6`A2 e−2Ep /T + e−4Ep /T ,              (2.45)

and similarly for SR in higher representations (with more complicated expressions). This
suggests treating the confinement dynamics and the interaction between the quark and
gluon sectors not in terms of a single traced Polyakov loop in the fundamental represen-
tation but in terms of eigenvalues of the Polyakov-loop matrix, which goes in the line of

                                                          – 13 –
the matrix-model approach [22, 81, 91, 92]. Moreover, for the convenience of studying
the bubble nucleation, some method needs to be introduced to reduce the multi-variable
problem to the tunnelling in a single dimension [22]. The extension of the current work to
these cases will be left for future study.
                                       h       i
     The medium potential Vmedium hψ̄ψi, `, `∗ does not contain ultraviolet (UV) diver-
gence and there are different procedures on the market regarding the regularization of this
contribution, even within a 3D momentum cutoff framework. For example, in [60], no mo-
mentum cutoff is imposed on the medium potential and the 3D momentum is integrated
to infinity. On the other hand, in [93], a sharp 3D momentum cutoff has been employed

                                                                                                                   JHEP01(2022)003
everywhere, including the medium potential. The choice is motivated by the authors’
wish to describe certain mesonic properties. When it comes to quarks in higher repre-
sentations, [90] regulates the medium potential by introducing a momentum-dependent
four-fermion coupling

                                             p |) = GS θ(Λ − |~
                                        GS (|~                p |) ,                                    (2.46)

which implies that for three-momentum larger than Λ, the medium potential is not set to
zero, but rather computed as if the quarks have zero constituent mass. It is found in [90]
that such a regularization treatment is needed to obtain clearly separated confinement and
chiral phase transition in the case of adjoint quarks.7

2.6    Model parameters and observables
Apart from the coefficients in the Polyakov-loop potential, the PNJL models we are consid-
ering have only two parameters (GS and Λ) in the 3G1 and 3S1 cases, and three parameters
(GS , GD , and Λ) in the 3F3 case. In principle, these parameters should be determined
from observables (meson masses, decay constants) measured from experiments or predicted
in lattice calculations. However, because we work in the chiral limit, even in the 3F3 case
it is difficult to determine the parameters precisely. In [19], four benchmark points were
chosen to study the 3F3 model in the chiral limit. We deem it reasonable in the 3F3 case
to use the parameters corresponding to physical real-world values as a reference and then
investigate variations away from the physical point by some amount. The values for such
a choice can be found in [57], see also [94]. In the 3G1 and 3S1 cases, however, there are
no clear guidelines to determine the parameter and we thus allow the parameters to vary
in a larger range.
     Nonetheless, we provide formulae for a set of observables to gain more physical insight
from the model parameters that we use, and also to facilitate future comparison of compu-
tations done in different approaches (because a meaningful comparison should be carried
out at the same value of observables). The set of observables that we consider include the
chiral condensate σ, the constituent quark mass M , the pion-decay constant fπ , and the
σ-meson mass mσ .
   7
    Interestingly, we observe that without this regularization treatment, it is possible to obtain a first-order
chiral phase transition for sufficient large GS while the confinement and chiral phase transition become
strongly correlated like in the fundamental case.

                                                    – 14 –
The chiral condensate σ is determined from the saddle point equation at zero temper-
ature and chemical potential

                                     ∂(Vcond + Vzero )
                                                       = 0,                            (2.47)
                                           ∂σ
which is just the gap equation of the corresponding NJL model. The constituent quark
mass M is then given by eqs. (2.26), (2.27), (2.28).
     The pion-decay constant is determined from the vacuum to one-pion axial-vector ma-
trix element, and we use the normalization in [55]. In the 3D momentum cutoff scheme, it
is given by

                                                                                                 JHEP01(2022)003
                                s            s
                                    dim(R)                 Λ       Λ
                       fπ = M                    arcsinh     −√        .               (2.48)
                                      2π 2                 M    Λ + M2
                                                                 2

Finally, the σ-meson mass mσ is the root of the 1PI σ-σ 2-point function Γσσ whose
expressions are lengthy and thus given in appendix A. It turns out that in the 3G1 and
3S1 cases the σ-meson mass is simply twice the constituent quark mass.

2.7   Comments on universality
We have collected various pieces of the PNJL grand potential for the three models of our
interest. Before moving to the exploration of the phase transitions based on the expressions
for the grand potential, let us comment on the relation between the universality argument
and our analysis. There is a nice summary of the logic of the universality argument in [95]:

  1. One first assumes the phase transition is continuous. The asymptotic critical be-
     haviour must be associated with a 3D universality class with the same symmetry
     breaking pattern as the original theory.

  2. The existence of such a 3D universality class can be investigated by considering the
     most general Landau-Ginzburg-Wilson Φ4 theory compatible with the given symme-
     try breaking pattern.

  3. The critical behaviour at a continuous transition is determined by the fixed points
     (FPs) of the renormalization group (RG) flow: the absence of a stable FP generally
     implies first-order transitions. However, if a stable FP exists, the phase transition
     can be of second-order or first-order (if the theory is outside the attractive domain
     of the FP).

RG predictions for the type of chiral phase transitions in SU(3) QCD theories with quarks
in the complex or real representations have also been summarized in [95]. The universality
analysis can also be carried out for the confinement phase transition, see [50] for a summary.
Among the three models we are considering, 3F3 is predicted to exhibit a first-order chiral
phase transition. On the other hand, for 3G1 and 3S1, no definite prediction can be made.
It is therefore well-motivated to explore what type of phase transitions these models exhibit
using an effective theory approach like PNJL.

                                             – 15 –
For quarks in the fundamental representation, the universality argument predicts a
first-order phase transition also for a number of Dirac flavours larger than 3 (but below the
lower boundary of the conformal window). On the other hand, our study suggests that for
Nf = 4 the PNJL approach does not seem to exhibit a first-order chiral phase transition
even if the model parameters are allowed to vary in a large range. This does not mean the
universality argument is invalid. Rather, it is likely that this points to the possibility that
the PNJL approach fails to model the phase-transition dynamics faithfully in such a case.
This motivates further modelling of the phase transition using alternative effective theories
in order to deliver a first-order chiral phase transition compatible with the prediction of
universality.

                                                                                                  JHEP01(2022)003
3     Confinement and chiral phase transitions

In this section, we discuss the nature of the confinement and chiral phase transition for
the models studied in this work. We start with some cosmological considerations and the
discussion of order parameters, followed by a generic review of the bubble nucleation and
the resulting GW spectrum.

3.1    Cosmological considerations
We briefly discuss here the cosmological constraints on the dark pions in our model. This
can be divided into the two cases of massless and massive dark pions.

Massless dark pions. The dark pions play the role of dark radiation which is strongly
constrained by the cosmic microwave background (CMB). Following the nice discussion
in [96], there are two viable cases:

    1. The dark sector and visible sector are thermalized in the very early universe but
       decouple prior to the electroweak scale Tew . In addition, the chiral phase transition
       should happen even before the decoupling. In this case, for the SU(3) gauge group
       with fermions in the fundamental and two-index symmetric representation, only 1 ≤
       Nf ≤ 3 Dirac flavours are viable while in the adjoint representation, only one Dirac
       flavour is viable. Thus our choice of Nf = 3 and Nf = 1 in respectively fundamental
       and adjoint (two-index symmetric) representations is valid.

    2. The dark sector and visible sector never thermalize. In this case, the key parameter
       from the CMB constraint i.e. the ratio between the hidden and visible sector tem-
       perature during the CMB epoch Td (tCMB ) /Tv (tCMB ) can be made arbitrarily small
       and thus avoid the constraints from CMB. Note that the GW signal is suppressed if
       the dark temperature is much colder than the visible temperature [97, 98] and thus
       we do not consider this case.

Massive dark pions. This scenario is less constrained and can be achieved by adding
small explicit quark masses. We assume the dark gauge sector is in thermal equilibrium
with the SM at early times before the Big-Bang Nucleosynthesis (BBN) and that the
pseudo-Nambu-Goldstone bosons (pNGBs) have decay channels into lighter SM particles.

                                            – 16 –
As long as they have a mass larger than a few MeVs and can decay before BBN, these
pNGBs do not cause conflicts with cosmological and laboratory constraints [99], which we
explain in more detail in appendix B.
     In this work, we consider the simplest case where at the phase transition the dark and
visible sectors are in thermal equilibrium, Td = Tv . This implies that the dark pions can
be massless for Tc > Tew , while they have to be massive for Tc < Tew .

3.2   Order parameters

In this section, we focus on the interplay between chiral and confinement phase transitions

                                                                                                    JHEP01(2022)003
for the models studied here, see table 1. The order parameter of the confinement phase
transition is the Polyakov-loop expectation value, while the chiral condensate is the order
parameter for the chiral phase transition. Fermions in the fundamental and two-index
symmetric representation explicitly break the Z3 centre symmetry of SU(3) and thus the
Polyakov loop is no longer a rigorous order parameter for the confinement phase transition.
Nevertheless, the Polyakov loop can still serve as an indicator of a crossover between
confinement and deconfinement [61, 100].
     We display the expectation values of the Polyakov loop h`i and the chiral condensate
σ (or equivalently of the constituent quark mass M ) as a function temperature for each
case, see figure 1, figure 2 and figure 3. The chiral condensate and the constituent mass are
normalised to their respective values at vanishing temperature. We display the polynomial
and logarithmic fitting of the effective Polyakov-loop potential by eqs. (2.17), and (2.15).
In the pure-glue case, both potentials have the property that h`i → 1 for T → ∞, however,
with the addition of the medium potential eq. (2.32) this property is lost in the polynomial
case. In principle, one would need to refit the coefficients to the lattice data with the refined
constraint of h`i → 1 for T → ∞ that includes the properties in the medium potential. We
refrain from doing so since this effect only becomes relevant at large T and is not relevant
for the dynamics of the phase transition. For the purpose of the figures, we implement the
constraint h`i ≤ 1 by hand. We also emphasise that the logarithmic potential naturally
implements the constraint h`i → 1 for T → ∞ and therefore might be better suited for the
considerations in this work.

Fundamental representation SU(3) with Nf = 3 (figure 1). The chiral condensate
has a discontinuity at the critical temperature and thus we have a first-order chiral phase
transition. The Polyakov loop expectation value undergoes a cross over (there is a small
discontinuity at Tc due to the discontinuity in σ). The confinement crossover happens
roughly at the same temperature as the first-order chiral phase transition.

Adjoint Representation SU(3) with Nf = 1 (figure 2). The fermions in the adjoint
representation do not break the Z3 centre symmetry and thus the Polyakov loop expectation
value remains a good order parameter for the confinement phase transition. We find a first-
order confinement phase transition, while the chiral phase transition is of second order and
happens at much larger temperatures.

                                             – 17 –
1

                                                                   h`ipoly
                                                                    h`ilog
               0.5                                                σpoly /σ0
                                                                  σlog /σ0

                                                                                                     JHEP01(2022)003
                 0
                      0.6     0.8      1       1.2 1.4        1.6     1.8       2
                                               T /Tc
Figure 1. Fundamental representation: expectation value of the Polyakov loop h`i and the chiral
condensate σ as a function of temperature. The latter is normalised to its value at vanishing
temperature σ0 . We present the polynomial and logarithmic fitting of the Polyakov-loop potential.

                 1

               0.5
                                            h`ipoly
                                             h`ilog
                                           Mpoly /M0
                                           Mlog /M0
                 0
                        1        2         3        4        5        6         7
                                                T /Tc
Figure 2. Adjoint representation: expectation value of the Polyakov loop h`i and the constituent
quark mass M as a function of temperature. The latter is normalised to its value at vanishing
temperature M0 . We present the polynomial and logarithmic fitting of the Polyakov-loop potential.

Two-index symmetric representation: SU(3) with Nf = 1 (figure 3). The two-
index symmetric representation is a very interesting case. In principle, the centre symmetry
is explicitly broken by the fermions and thus there should be no confinement phase tran-
sition. However, it turns out that the centre symmetry is only weakly broken [90]. The
amount of symmetry breaking is characterised by 1/M where M is the constituent quark

                                               – 18 –
1

               0.5
                                                 h`ipoly
                                                  h`ilog
                                                Mpoly /M0

                                                                                                      JHEP01(2022)003
                                                Mlog /M0
                  0
                         1          2                    3             4             5    6
                                                         T /Tc
Figure 3. Two-index symmetric representation: expectation value of the Polyakov loop h`i and
the constituent quark mass M as a function of temperature. The latter is normalised to its value at
vanishing temperature M0 . We present the polynomial and logarithmic fitting of the Polyakov-loop
potential.

mass. The latter is rather large in the two-index symmetric representation, see table 3, and
consequently, the centre symmetry is only softly broken. As in the adjoint case, the chiral
phase transition is of second-order and happens at larger temperatures. Therefore the
centre symmetry is almost restored at Tc , and we observe a first-order confinement phase
transition. The small negative dip of the Polyakov loop expectation value is precisely due
to the breaking of the centre symmetry induced via the medium potential eq. (2.32).

3.3   Bubble nucleation
In case of a first-order phase transition, the transition occurs via bubble nucleation and
it is essential for the understanding of the dynamics to compute the nucleation rate. The
tunnelling rate due to thermal fluctuations per unit volume as a function of the temperature
from the metastable vacuum to the stable one is suppressed by the three-dimensional
Euclidean action S3 (T ) [101–104]
                                                              3/2
                                                    S3 (T )
                                                
                               Γ(T ) = T    4
                                                                     e−S3 (T )/T .            (3.1)
                                                     2πT
The three-dimensional Euclidean action reads
                                        Z ∞              "      2                  #
                                                     2    1 dρ
                         S3 (T ) = 4π       dr r                       + Veff (ρ, T ) ,       (3.2)
                                        0                 2 dr
where ρ denotes a generic scalar field with mass dimension one, [ρ] = 1, and Veff denotes
its effective potential. In our case, the effective potential depends on two scalar fields, the
Polyakov loop ` and the chiral condensate σ. Which field takes the leading role depends
on whether we have a first-order confinement or chiral phase transition and therefore we
discuss them separately.

                                                    – 19 –
Confinement phase transition. The phase transition is described by the Polyakov
loop ` and it is a first-order phase transition in the adjoint and two-index symmetric case,
see figure 2 and figure 3. In both cases, the second-order chiral phase transition is at
significantly higher temperatures and has already been completed. Therefore, we can work
in the approximation that σ is constant. Note also that ` is dimensionless while ρ in
eq. (3.2) has mass dimension one. We therefore rewrite the scalar field as ρ = ` T and
convert the radius into a dimensionless quantity r0 = r T . Thus, the action becomes
                                      Z ∞                    "              2                      #
                                                     0 02      1       d`               0
                      S3 (T ) = 4πT            dr r                               +   Veff (`, T )       ,   (3.3)
                                          0                    2       dr0

                                                                                                                     JHEP01(2022)003
                                              0 (`, T ) = V (`, T )/T 4 is dimensionless. The
which has the same form as eq. (3.2). Here, Veff           eff
bubble profile (instanton solution) is obtained by solving the equation of motion of the
action in eq. (3.3)

                           d2 `(r0 )   2 d`(r0 ) ∂Veff
                                                    0 (`, T )
                                     +          −             = 0,                                           (3.4)
                            dr02       r0 dr0        ∂`
with the associated boundary conditions

                   d`(r0 = 0, T )
                                  = 0,                                            lim `(r0 , T ) = 0 .       (3.5)
                        dr0                                                       r0 →0

To attain the solutions, we used the method of overshooting/undershooting and employ
the Python package CosmoTransitions [105].

Chiral phase transition. The chiral phase transition is described by the chiral conden-
sate σ, see eqs. (2.21), (2.23). In the three models studied here, we only find a first-order
chiral phase transition in the fundamental case, see figure 1. In order to have a field with
mass dimension one, we define

                                                σ̄ ≡ −4GS σ .                                                (3.6)

We work in the mean-field approximation where we evaluate the Polyakov loop ` for given
values of σ̄ and T at the minimum of the effective potential. Thus the potential becomes
a function of only (σ̄, T ), Veff (σ̄, T ) = Veff (σ̄, T, `min (σ̄, T )).
     Since σ̄ is not a fundamental field, we have to include its wave-function renormalization
Zσ , see appendix A for more details. In figure 4, we display the wave-function renormal-
ization as a function of the chiral condensate and the temperature. The three-dimensional
Euclidean action is slightly modified [19]
                                                         "                                           #
                                                             Zσ−1 dσ̄
                                     Z ∞                                    2
                                                     2
                      S3 (T ) = 4π            dr r                                + Veff (σ̄, T ) .          (3.7)
                                      0                       2   dr

The bubble profile is obtained by solving the equation of motion of the action in eq. (3.7)
and is given by
                       d2 σ̄ 2 dσ̄ 1 ∂ log Zσ dσ̄ 2          ∂Veff
                                                    
                           2
                             +      −                   = Zσ       ,                  (3.8)
                        dr     r dr   2 ∂ σ̄      dr          ∂ σ̄

                                                         – 20 –
0.4
                                                                                           0.5Tc
                                                                                           0.75Tc
                                                                                             Tc
               Zσ−1 (σ, T )                                                                1.5Tc
                              0.2                                                           2Tc

                                                                                                                 JHEP01(2022)003
                               0

                                    0        0.5            1                        1.5            2
                                                           σ/σ0

Figure 4. Wave-function renormalization Zσ−1 for different temperatures as a function of the chiral
condensate σ normalised to its value at zero temperature σ0 .

with the associated boundary conditions
                              dσ̄(r = 0, T )
                                             = 0,                             lim σ̄(r, T ) = 0 .        (3.9)
                                    dr                                    r→∞

For Zσ = 1, eq. (3.8) simplifies to eq. (3.4). We use again the overshooting/undershooting
method and employ the Python package CosmoTransitions [105] with a modified equation
of motion. We substitute the solved bubble profile σ̄(r, T ) into the three-dimensional
Euclidean action eq. (3.7) and, after integrating over r, S3 depends only on T .

3.4     Gravitational-wave parameters
3.4.1    Inverse duration time
An important parameter for determining the GW spectrum is the rate at which the phase
transition completes. For sufficiently fast phase transitions, the decay rate can be approx-
imated by

                                              Γ(T ) ≈ Γ(t∗ )eβ(t−t∗ ) ,                                 (3.10)

where t∗ is a characteristic time scale for the production of GWs to be specified below.
The inverse duration time then follows as
                                                       d S3 (T )
                                                 β=−                      .                             (3.11)
                                                       dt T        t=t∗

The dimensionless version β̃ is defined relative to the Hubble parameter H∗ at the charac-
teristic time t∗
                                                 β      d S3 (T )
                                          β̃ =      =T                           ,                      (3.12)
                                                 H∗    dT T           T =T∗

                                                       – 21 –
where we used that dT /dt = −H(T )T . Note that here we assumed that the temperature
in the hidden and visible sectors are the same, Td = Tv .
     The phase-transition temperature T∗ is often identified with the nucleation temperature
Tn , which is defined as the temperature at which the rate of bubble nucleation per Hubble
volume and time is approximately one, i.e. Γ/H 4 ∼ O(1). More accurately one can use the
percolation temperature Tp , which is defined as the temperature at which the probability
to have the false vacuum is about 0.7. For very fast phase transitions, as in our case, the
nucleation and percolation temperature are almost identical Tp . Tn . However, even a
small change in the temperature leads to an exponential change in the vacuum decay rate
Γ, see eq. (3.10), and consequently, we use the percolation temperature throughout this

                                                                                                           JHEP01(2022)003
work. We write the false-vacuum probability as [106, 107]

                                            P (T ) = e−I(T ) ,                                    (3.13)

with the weight function [108]
                                                             Z T0                    00 ) 3
                                                                                         !
                                              Γ(T 0 )
                                  Z Tc
                             4π              0                            00 vw (T
                     I(T ) =              dT                         dT                       .   (3.14)
                              3       T      H(T 0 )T 04         T          H(T 00 )

The percolation temperature is defined by I(Tp ) = 0.34, corresponding to P (Tp ) = 0.7 [109].
Using T∗ = Tp in eq. (3.12) yields the dimensionless inverse duration time. We will see
that all phase transitions considered here have very fast rates, β̃ ∼ O(104 ).

3.4.2   Energy budget
We define the strength parameter α from the trace of the energy-momentum tensor θ
weighted by the enthalpy
                                           1 ∆θ   1 ∆e − 3∆p
                                  α=            =            ,                                    (3.15)
                                           3 w+   3    w+

where ∆X = X (+) −X (−) for X = (θ, e, p) and (+) denotes the meta-stable phase (outside
of the bubble) while (−) denotes the stable phase (inside of the bubble). The relations
between enthalpy w, pressure p, and energy e are given by
                             ∂p                                              ∂p
                     w=           ,                                  e=           − p.            (3.16)
                           ∂ ln T                                          ∂ ln T
These are hydrodynamic quantities and we work in the approximation where do not solve
the hydrodynamic equations but instead extract them from the effective potential with
                                                           (±)
                                            p(±) = −Veff .                                        (3.17)

This treatment should work well for the phase transitions considered here, see [110–112].
With eqs. (3.16), (3.17), α is given by

                                            1 4∆Veff − T ∂∆V∂T
                                                               eff
                                   α=                   (+)
                                                                   .                              (3.18)
                                            3        ∂Veff
                                                 −T ∂T

                                                 – 22 –
You can also read